Ci Song

I am a Ph.D. student under the supervision of Prof. Tobias Oechtering and the co-supervision of Prof. Magnus Jansson. I obtained my Master's degree in Information and Network Engineering in 2024 from KTH, Sweden, and my Bachelor's degree in Information Engineering in 2022 from Southeast University, China.

My research focuses on trustworthy digital twins for smart buildings. My research interests include: representativeness, privacy and bias of models; privacy-preserving data publishing; signal processing; pattern recognition and machine learning.

A digital twin[1] is the virtual counterpart of a physical entity or process. It is a living and evolving model that follows the lifecycle of its physical twin. It continuously predicts future states (e.g., defects, damages, failures). It allows simulating and testing novel configurations. It can monitor, control, and optimize its physical twin (e.g., management, fault detection).

To preserve privacy, information-theoretic approaches typically inject noise into data before disclosure. However, noise injection may significantly degrade data utility. Motivated by robustness considerations in the EU AI Act, [2] studies privacy mechanism design under a worst-case utility criterion. We show that Pointwise Maximal Leakage (PML) enables mechanisms that prevent certain highly undesirable outputs while preserving formal privacy guarantees. This paper mainly focuses on the theoretical foundation and demonstrates the idea through examples, while practical implementations are part of ongoing work.

[1] B. R. Barricelli, E. Casiraghi and D. Fogli, "A Survey on Digital Twin: Definitions, Characteristics, Applications, and Design Implications," inIEEE Access, vol. 7, pp. 167653-167671, 2019, doi: 10.1109/ACCESS.2019.2953499.

[2] Ci Song and Tobias J. Oechtering, “Worst-Case Utility Privacy Mechanism via Pointwise Maximal Leakage,” arXiv preprint arXiv:2605.19474, 2026. Accepted to EUSIPCO 2026.